Explainable AI, Causal Graphs and an Application to Insurance Ratings

Holger Bartel, the founder of RealRate, shows how AI can be made explainable even for a small amount of data, like annual report data.


A structural neural network restricts the usually fully connected layers. Thus, the model is explainable by design.

The methodology makes use of graph theory and derivatives. A hybrid model is used, consisting of an expert system, defining the data relationship in terms of mathematical equations, and the usual supervised learning to determine predefined model parameters.

All parameters, and even every single node in the network, are economically interpretable. This approach is applied to German health insurers to create a financial strength rating.

The results are explained with a focus on the causal graph, showing the strengths and weaknesses of each individual insurer. 2023-09-10.


Original Actuview source:


The video is in German.


YouTube link:



RealRate is the first AI rating agency. We have offices in Santa Clara and Berlin. We cover German insurers and more than 20 US industries.